Therapeutic Assessment Evaluation on TheraJudge Evaluation Set
0.949Avg ICCTheraJudge
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TheraJudgeGeneration/Iteration=Gen 16, Optimization=GRPO2026.06 | 0.949 | 1 | 87.9 | 96 | 98.9 | 93.2 | — | |
| TheraJudgeGeneration/Iteration=Gen 42026.06 | 0.924 | 2 | — | — | — | — | — | |
| TheraJudgeGeneration/Iteration=Gen 82026.06 | 0.9 | 3 | — | — | — | — | — | |
| Claude-3.7Full Model Name=Claude-3.7-Sonnet2026.06 | 0.83 | 4 | 68.5 | 73 | — | — | — | |
| Qwen-SFTBase Model=Qwen-2.5-7B, Training Stage=Supervised Fine-Tuning2026.06 | 0.827 | 5 | — | — | — | — | — | |
| GPT-4o2026.06 | 0.739 | 6 | 48 | — | — | — | — | |
| o4-mini2026.06 | 0.727 | 7 | 25.9 | — | — | — | — | |
| Qwen-Zero-shotBase Model=Qwen-2.5-7B, Training Stage=Zero-shot2026.06 | 0.623 | 8 | 14.5 | 51.9 | 65 | 61.6 | — | |
| Gemini-2.5Full Model Name=Gemini-2.5-Flash2026.06 | 0.621 | 9 | 37.7 | 30.6 | — | — | — |